[AutoBE] We made AI-friendly Compilers for Vibe Coding, achieving 100% build success

Preface

GitHub Repository  | Generation Result 

Wrtn Technologies introduces AutoBE, an open-source vibe coding agent that automatically generates backend applications. The standout feature is guaranteed 100% compilation success through proprietary compiler systems that validate AI-generated Abstract Syntax Trees and provide detailed correction feedback.

AI Function Calling is a technology where AI generates structured data according to predefined function schemas, producing JSON data adhering to specific types and formats.

Waterfall Compiler System

Outline

AutoBE employs a five-phase waterfall model with dedicated agents: Analyze (requirements), Prisma (database schema), Interface (API design), Test (E2E testing), and Realize (implementation). Each phase’s output undergoes validation through specialized compilers — a Prisma compiler, OpenAPI validator, and TypeScript compiler — ensuring sequential quality assurance.

Prisma DB Schema Compiler

The database compiler creates AutoBePrisma.IFile AST structures through AI function calling, then validates them for logical and type errors. Detected issues — including duplication, circular references, non-existent references, index errors, type mismatches, and naming conflicts — are returned to the AI with detailed explanations for correction.

Once validated, the system generates Prisma schemas with detailed comments, Entity Relationship Diagrams, and accompanying documentation via the prisma-markdown tool.

OpenAPI Document Compiler

The API compiler similarly constructs AutoBeOpenApi.IDocument through AI function calling and validates it comprehensively. After successful validation, it converts to official OpenAPI v3.1 specifications and generates TypeScript/NestJS source code with mock endpoints using typia.random<T>().

All generated code includes detailed JSDoc comments describing endpoints, parameters, and return values.

E2E Test Function Compiler

For TypeScript/NestJS stacks, the test compiler differs slightly: the AI devises test scenarios, writes draft code, self-reviews, and produces final implementations. The TypeScript compiler validates this code; compilation failures trigger detailed feedback for AI correction.

Generated tests include comprehensive comments explaining scenarios, API call sequences, and expected verification outcomes.

Additional Resources

TypeScript Compiler

AutoBE embeds the TypeScript compiler for validating both AI-generated and internally-generated TypeScript code.

AI Function Calling Compiler

The project utilizes the typia  library — a TypeScript compiler plugin — to automatically generate AI function calling schemas from TypeScript source code. This solves the complexity problem: compiler AST structures are too intricate for manual JSON schema definition, especially given inconsistent JSON schema specifications across AI vendors (OpenAI, Gemini, and Claude each use different approaches).

typia.llm.application<Class, Model>() generates both function calling schemas and validator functions for type validation feedback.

Backend Stack

TypeScript / NestJS / Prisma

AutoBE currently supports this combination exclusively. TypeScript was chosen for its open-source, extensible compiler with a robust type system. NestJS, paired with Nestia (which auto-generates client SDKs), enables type-safe E2E testing. Prisma provides structured schema definitions and type-safe query generation.

Other Languages and Frameworks

The architecture theoretically supports other languages, but language-specific expansion requires substantial investment in understanding each language’s type system, compiler characteristics, and framework structure while maintaining the 100% compilation success guarantee.

Development Status and Roadmap

AutoBE is in alpha with incomplete features:

• Realize agent incomplete: The implementation phase isn’t finished; beta release scheduled for August 31, 2025
• Unoptimized prompts: Generated code compiles and functions correctly but may not match intended functionality
• No RAG optimization: Current API token consumption approximates $30 for applications with 200 endpoints using GPT-4.1; future optimization will reduce this

Conclusion

AutoBE distinguishes itself by combining compiler technology with AI, enabling structurally safe code generation rather than relying on text-based approaches. The compiler-validates-AI-output cycle attempts fundamental quality improvement beyond convenience enhancement.

The project remains in alpha development, with future IDE plugin integration planned alongside the v1 release.

Appendix: AutoBE vs AWS Kiro

AWS Kiro represents a significant AI development tools milestone through IDE plugin implementation. The comparison reveals AutoBE’s early-stage status: Kiro has achieved market maturity with proven adoption, while AutoBE remains in alpha with incomplete realize agent and unoptimized prompts.

Rather than competition, the team views this as learning opportunity — studying Kiro’s user experience design, IDE integration patterns, and workflow optimization while leveraging AutoBE’s compiler-based architecture and guaranteed compilation success.

The planned IDE plugin for VSCode will integrate AutoBE’s waterfall compiler system, enabling in-IDE backend application generation with real-time compiler feedback.